Goldman Sachs: The $27 Trillion AI Stock Market Is Shifting from a Valuation Bubble to an Earnings Bubble

From Valuation Bubble to Earnings Bubble

The most important conceptual shift in Goldman's June 2026 analysis is the evolution of the primary risk. Upward revisions to earnings have temporarily alleviated classic valuation concerns. However, the bank now warns that the real danger is an "earnings bubble": if earnings growth disappoints, or if the cyclical investment peak arrives, the market could shift from worrying about high multiples to confronting the fact that earnings themselves are unsustainable. The implication is that investor sensitivity to narrative changes is rising, and prices have already embedded a great deal of optimistic expectations.

The $5.3 Trillion Capex Supercycle

Goldman Sachs estimates that between 2025 and 2030, hyperscale cloud firms will spend a total of $5.3 trillion on AI and data center infrastructure. The bank describes this as an unprecedented capital expenditure supercycle, and it warns that this wave of spending is not only unsustainable but is actively eroding the financial returns of the largest technology companies. The bank's analysts explicitly state that consensus expectations for 2027 hyperscaler capex are "too conservative"—their own estimates suggest spending could reach roughly $1.1 trillion in 2027, compared to the roughly $920 billion expected by Wall Street, with a bullish scenario reaching $1.4 trillion.

Underlying Weakness in the Non-AI Economy

On June 2, 2026, Goldman Sachs trader Lee Coppersmith issued a stark warning: while the index-level rally appears smooth, the underlying dynamics are becoming "increasingly unsettling." He noted that market bets on AI have evolved from being fundamentally driven to a self-referential cycle amplified by market structure itself. Positions are more crowded, leverage is higher, and concentration is greater—yet the cost investors pay to protect against downside risk has fallen to historic lows. This dynamic masks broader economic softness, creating a fragile foundation for the rally.

Semiconductor Concentration: The New Fault Line

The semiconductor sector has captured an outsized share of AI profits, and Goldman Sachs describes this concentration as unsustainable. Jim Covello, head of research at Goldman Sachs, continues to argue that 95% of enterprise organizations are yielding zero ROI from AI, and that the profit concentration in chipmakers is a structural fragility. Covello has pointed out that in many ways, companies are losing more money today implementing this technology than they were two years ago. Despite some hedge fund selling in semiconductor stocks recently, exposure to AI stocks within Goldman's TMT tracking basket remains close to all-time peaks. The bank has proposed a relative-value trading strategy: go long on hyperscale cloud providers and underweight semiconductors, arguing that the market has yet to price in the risk of a capex slowdown hitting chipmakers hardest.

The Verdict: Stay Invested, but Hedge

Goldman Sachs has not recommended exiting AI stocks. Instead, the bank's core message is to stay invested but hedge downside risks. The bank characterized the $1.3 trillion AI selloff in late May/early June 2026 as a "stress test" rather than a lasting trend shift, and its composite sentiment gauge showed that positioning was not yet dangerously crowded. At the same time, Goldman Sachs raised its year-end S&P 500 target, while separately flagging that excessive speculative activity and high valuations are the two primary risks that have historically ended bull markets.

The bank's overall stance is "constructive on equities" with a preference for maintaining AI exposure, but with heightened awareness of concentration risk, rising volatility, and the need for protection against a potential capex slowdown or earnings disappointment. In effect, Goldman Sachs is telling investors to prepare for a market where the narrative can turn faster than the fundamentals—and where the biggest risk may not be being wrong about AI, but being overexposed when the cycle turns.